Software Pipelining in the LLVM Compiler

This thesis discusses a design and implementation of the Software Pipelining, a optimization technique of loops in a program, which tries to exploit instruction-level parallelism. It is achieved by scheduling instructions in a way to overlap iterations of the loop and therefore execute them in a pipeline. This way optimization speeds up the final program. There is a detailed description of design and implementation of Swing Modulo Scheduling algorithm, an effective and efficient method for finding near-optimal plans for software-pipelined loops. This work has been done as a part of a larger project, the development of Codasip Framework. Part of this framework is the retargetable C compiler based on compiler architecture LLVM, in which this work is implemented